What’s Next for Microservices? Highlights from the Beijing Cloud Native Meetup

Next‑Generation Microservice Architecture

Li Yanlin (founder of Nacos & Higress, senior technologist at Alibaba Cloud) described the parallel evolution of multiple microservice stacks such as Spring Cloud Alibaba, Dubbo, and Envoy. He presented a "microservice panorama" that outlines how services, service mesh, and API gateways interoperate, and he highlighted design principles for building secure, highly‑available microservice systems.

Static Compilation with GraalVM for Spring Cloud Alibaba

With the release of Spring Boot 3.0, GraalVM native‑image support enables Java applications to be compiled ahead‑of‑time into native binaries. This eliminates the JVM warm‑up phase, reduces cold‑start latency from seconds to sub‑second levels, and cuts runtime memory usage by up to 70 %.

Rao Zihao (Spring Cloud Alibaba community lead) explained how the Spring Cloud Alibaba libraries have been updated to be compatible with Spring Boot 3.0 and GraalVM. Key steps include:

Adding --initialize-at-build-time flags for third‑party dependencies.

Providing native-image.properties files to configure reflection and resource bundles.

Using the spring-graalvm-native plugin to generate Docker images that embed the native binary.

These changes allow Spring Cloud Alibaba‑based microservices to run as lightweight native containers, which is especially beneficial in serverless or edge‑cloud scenarios.

Dubbo as a Preferred Runtime for gRPC‑Based Services

Chen Youwei (Apache Dubbo committer) compared Dubbo with raw gRPC. While gRPC defines the wire protocol, it does not provide a programming model, integration with Spring, or traffic‑control primitives. Dubbo adds:

Declarative service interfaces with annotations for versioning and routing.

Built‑in support for service discovery (e.g., Nacos, Zookeeper).

Fine‑grained traffic management (rate limiting, circuit breaking) via the Dubbo‑SPI.

By wrapping gRPC calls, Dubbo lets developers write standard Java interfaces while still benefiting from gRPC’s high‑performance binary protocol.

Higress: Cloud‑Native Microservice Gateway

Zhang Tianyi (Higress maintainer) traced the evolution from traditional API gateways to cloud‑native gateways. Higress distinguishes itself with:

Dynamic configuration hot‑reload without pod restarts.

Extensible filter chain based on Envoy’s WASM plugins.

Built‑in observability (metrics, tracing) integrated with OpenTelemetry.

Transparent RPC protocol conversion (e.g., HTTP↔gRPC) to unify service access.

These capabilities simplify the deployment of zero‑downtime API routing and service mesh ingress.

Sentinel 2.0 Service Governance

Pan Shengwei (OpenSergo maintainer) introduced Sentinel 2.0, which now implements the OpenSergo standard for cloud‑native governance. Core features include:

Universal flow control and circuit breaking that work across Dubbo, Spring Cloud, and gRPC.

Declarative policy definition via YAML or OpenSergo’s CRD.

Runtime policy propagation using a sidecar model, enabling zero‑downtime updates.

A live demo showed Sentinel 2.0 enforcing rate limits on a Dubbo service according to an OpenSergo policy.

Seata: Distributed Transaction Coordination

Ji Min (Seata founder) explained the challenge of maintaining data consistency across microservices. Seata’s architecture consists of:

Transaction Manager (TM) that initiates global transactions.

Resource Manager (RM) that registers branch transactions with underlying databases.

Session Manager (SM) that persists transaction state for recovery.

Recent community progress adds support for TCC, AT, and Saga models, and the roadmap includes tighter integration with Spring Cloud Alibaba and improved observability hooks.

Fluid: Extending Compute Boundaries for Machine Learning

Che Yang (Fluid co‑founder) described how CNCF Fluid decouples storage from compute in hybrid‑cloud environments. Fluid provides:

Dataset and model abstractions that can be mounted as CSI volumes.

Automatic data locality optimization, moving data closer to GPU‑enabled nodes.

Cost‑aware scheduling that selects on‑prem or cloud storage based on policy.

This approach reduces data transfer overhead and improves GPU utilization for ML workloads.

Open Cluster Management (OCM): Multi‑Cluster Toolbox

Qiu Jian (OCM lead) outlined OCM’s primitive resources—ManagedCluster, Placement, and AddOn—that simplify multi‑cluster lifecycle management. OCM integrates with:

ArgoCD for Git‑Ops continuous delivery across clusters.

Submariner for cross‑cluster networking.

Cluster‑wide RBAC propagation via OpenShift‑style policies.

Users can extend OCM by developing custom AddOns that expose additional cluster capabilities.

Koordinator: Efficient Container Resource Scheduling

Zeng Fansong (Koordinator maintainer) showed how Koordinator introduces a “resource‑leveling” scheduler that classifies workloads into latency‑sensitive (microservices, front‑ends) and batch‑heavy (AI, video transcoding). The scheduler:

Applies priority‑based preemption to protect latency‑critical pods.

Performs load‑leveling across nodes to smooth CPU/GPU spikes.

Reduces overall cluster cost by up to 30 % in mixed‑workload scenarios.

Real‑Time Observability for Kubernetes Applications

Zeng Qingguo (KubeVela maintainer) demonstrated a monitoring stack built with KubeVela, ArgoCD, and Helm. The workflow collects metrics, logs, and traces via OpenTelemetry, aggregates them in Prometheus/Grafana, and visualizes health dashboards. The stack also supports automated alerting based on SLA thresholds.

OpenKruise: Enhanced Pod Lifecycle Management

Zhao Mingshan (OpenKruise maintainer) explained extensions beyond native Kubernetes, including:

GracefulPodLifecycle that delays termination until in‑flight requests finish.

Advanced LivenessProbe that aggregates health checks across containers.

SidecarSet for coordinated rollout of sidecar containers.

KubeVela + Kruise Rollout: Automated Canary Releases

Wang Yike (Alibaba Cloud senior engineer) showed how KubeVela’s plugin system can orchestrate a canary deployment:

# Define a KubeVela Application with a rollout strategy
apiVersion: core.oam.dev/v1beta1
kind: Application
metadata:
  name: my-service
spec:
  components:
  - name: backend
    type: webservice
    properties:
      image: myrepo/backend:{{version}}
    traits:
    - type: rollout
      properties:
        strategy: canary
        steps:
        - weight: 10
        - weight: 30
        - weight: 100

The rollout trait leverages Kruise Rollout to shift traffic gradually while monitoring health metrics, enabling fully unattended, observable releases.